Embed Size (px)
Citation preview
Institute for Empirical Research in Economics University of Zurich
Working Paper Series
ISSN 1424-0459
Working Paper No. 370
Risk Aversion
Pavlo R. Blavatskyy
April 2008
Risk Aversion Pavlo R. Blavatskyy
University of Zurich (IEW) Winterthurerstrasse 30
CH-8006 Zurich Switzerland
Phone: +41(0)446343586 Fax: +41(0)446344978
e-mail: [email protected]
April 2008
Abstract: Risk aversion is traditionally defined in the context of lotteries over monetary
payoffs. This paper extends the notion of risk aversion to a more general setup where
outcomes (consequences) may not be measurable in monetary terms and people may
have fuzzy preferences over lotteries, i.e. they may choose in a probabilistic manner.
The paper considers comparative risk aversion within neoclassical expected utility
theory, a constant error/tremble model and a strong utility model of probabilistic choice
(which includes the Fechner model and the Luce choice model as special cases). The
paper also provides a new definition of relative riskiness of lotteries.
Keywords: risk aversion, more risk averse than, riskiness, probabilistic choice,
expected utility theory, Fechner model, Luce choice model
JEL classification codes: D00, D80, D81
2
Risk Aversion Risk aversion is traditionally defined in the context of lotteries over monetary
payoffs (Pratt, 1964). However, one can also consider risk aversion when the outcomes
of risky lotteries may not be measurable in monetary terms. For example, people can be
risk averse or risk prone when driving their car, choosing a medical treatment, selecting
a holiday destination, deciding to marry or to divorce etc. This paper extends the notion
of risk aversion to decision problems where outcomes (consequences) may not be
measurable in monetary terms. Epstein (1999) defines uncertainty aversion when the
outcome set is arbitrary rather than Euclidean but people have a unique preference
ordering over uncertain alternatives.
Numerous empirical studies show that people generally have fuzzy preferences
over lotteries, i.e. they choose in a probabilistic manner (e.g. Camerer, 1989; Hey and
Orme, 1994; Loomes and Sugden, 1998). Therefore, this paper also extends the notion
of risk aversion to allow for the possibility of fuzzy preferences. Hilton (1989) and
Wilcox (2008) define risk aversion in the context of lotteries over monetary payoffs
when people choose probabilistically between lotteries.
The paper is organized as follows. Section 1 defines comparative risk aversion
in the context of an arbitrary outcome set. Section 2 considers the implications of this
definition for neoclassical expected utility theory. Section 3 extends the notion of risk
aversion to a more general setup where people have fuzzy preferences over lotteries.
Section 4 analyses risk aversion within two well-known models of probabilistic choice
(a constant error/tremble model and a strong utility model). Section 5 defines absolute
risk aversion and relative riskiness of lotteries. Section 6 concludes.
3
1. Comparative Risk Aversion Let X be a finite nonempty set of outcomes (consequences). We will treat X as
an arbitrary abstract set so that an element x∈X can be a monetary payoff, a consumption
bundle, a health state, marriage or divorce, birth of a child, the afterlife etc. A lottery
L:X→[0,1] is a probability distribution on X, i.e. it delivers an outcome x∈X with a
probability L(x) ∈ [0,1] and ∑x∈X L(x)=1. A degenerate lottery that yields one outcome
x∈X with probability one is denoted by (x,1). The set of all lotteries is denoted by ℒ.
In this and the next section we consider a “traditional” decision maker who has
a unique binary preference relation on ℒ. As customary, we will use the sign to
denote the asymmetric component of , and the sign ~ to denote the symmetric
component of . We will consider two individuals: an individual ♀ characterized by a
preference relation ♀ and an individual ♂ characterized by a preference relation ♂.
Definition 1 An individual ♀ is unambiguously more risk averse than an
individual ♂ if (x,1)♂L implies (x,1)♀L for all x∈X and all L∈ℒ and there exists at least
one degenerate lottery (x,1)∈ℒ and one lottery L∈ℒ such that (x,1) ~♂ L and (x,1) ♀ L.
According to Definition 1, a more risk averse individual weakly prefers a
degenerate lottery over another lottery whenever a less risk averse individual does so as
well. This definition of the more-risk-averse-than relation between individuals is very
general. Specifically, we do not require that lottery outcomes are measurable in real
numbers. We also do not require that individual preferences over lotteries are
represented by a specific decision theory (e.g. expected utility theory).
Definition 1 immediately implies the following result. If an individual ♀ is
more risk averse than an individual ♂, or vice versa, then (x,1) ♀ (y,1) if and only if
4
(x,1) ♂ (y,1) for all x,y∈X. This implication of our definition is quite intuitive. We can
unambiguously rank two individuals in terms of their risk preferences only if they have
identical preferences over riskless alternatives (degenerate lotteries). If the two
individuals do not have the same preferences in choice under certainty, one of them may
choose a specific degenerate lottery because it is her most preferred alternative and not
because she is averse to risk. Therefore, to have a meaningful concept of comparative
risk aversion, we need to consider individuals with identical preferences over the set of
riskless consequences.
2. Risk Aversion in Expected Utility Theory Let us now apply the concept of comparative risk aversion in the context of
expected utility theory (von Neumann and Morgenstern, 1944). In expected utility theory
there exists an utility function u:X→ that is unique up to a positive linear
transformation, such that
(1) S R if and only if ∑x∈X S(x)u(x) ≥ ∑x∈X R(x)u(x),
for any two lotteries S, R ∈ ℒ. Formula (1) simply states that a lottery S is weakly
preferred over a lottery R if and only if the expected utility of S is greater than or equal
to the expected utility of R.
As we already discussed above, for comparing risk aversion across individuals
we need to consider people with identical preferences over the set of riskless outcomes.
We will say that two individuals ♀ and ♂ have ordinally equivalent utility functions
when u♀(x)≥u♀(y) if and only if u♂(x)≥u♂(y) for any two outcomes x, y ∈ X.
Proposition 1 An expected utility maximizer ♀ with utility function u♀:X→
is more risk averse than an expected utility maximizer ♂ with ordinally equivalent
utility function u♂:X→ if and only if
5
(2) ( ) ( )( ) ( )
( ) ( )( ) ( )
u y u x u y u xu z u y u z u y
− −≥
− −♀ ♀ ♂ ♂
♀ ♀ ♂ ♂
,
for any x, y, z ∈ X such that u♀(x) < u♀(y) < u♀(z) and there exists at least one triple of
outcomes {x, y, z} ⊂ X for which inequality (2) holds with strict inequality.
Proof is presented in the Appendix.
Proposition 1 can be interpreted in the following way. We can use an index
I♀(x, y, z)≡(u♀(y)-u♀(x))/(u♀(z)-u♀(y)) to measure the risk aversion of an expected utility
maximizer ♀ in the context of outcomes x, y, z ∈ X such that u♀(x) < u♀(y) < u♀(z).
Similar to the Arrow-Pratt coefficient of absolute risk aversion for lotteries over
monetary payoffs, the index I♀(x, y, z) captures only local risk aversion. Therefore, for
an expected utility maximizer ♀ to be unambiguously more risk averse than an expected
utility maximizer ♂, we need to have I♀(x, y, z) ≥ I♂(x, y, z) for all triples of outcomes
{x, y, z} ⊂ X such that u♀(x) < u♀(y) < u♀(z). Note that an index I♀(x, y, z) is an adequate
measure of local risk aversion because it is invariant to positive linear transformations
of the utility function u♀:X→.
Figure 1 An expected utility maximizer ♀ is more risk averse than an expected utility maximizer ♂: illustration in the probability triangle
Probability of outcome x
1
Probability of outcome z
Slope ( ) ( )( ) ( )
u y u xu z u y
−−
♀ ♀
♀ ♀
10b) Indifference curves of an individual ♀
Probability of outcome x
1
Probability of outcome z
Slope ( ) ( )( ) ( )
u y u xu z u y
−−
♂ ♂
♂ ♂
10a) Indifference curves of an individual ♂
6
Proposition 1 can be conveniently illustrated inside the probability triangle
(e.g. Machina, 1982). When the outcome set X has only three elements (X={x, y, z}), the
set of all lotteries ℒ can be represented as a rectangular triangle. By convention, the
probability of the outcome that yields the lowest utility (x) is shown on the horizontal
axis and the probability of the outcome that yields the highest utility (z) is shown on the
vertical axis. Indifference curves inside the probability triangle represent the set of all
lotteries that yield the same expected utility. Specifically, indifference curves of an
expected utility maximizer with utility function u(.) are straight parallel lines with a
positive slope (u(y)-u(x))/(u(z)-u(y)). Thus, if an expected utility maximizer ♀ is more
risk averse than an expected utility maximizer ♂, the indifference curves of individual
♀ have a greater slope inside the probability triangle (indifference curves are steeper).
3. Probabilistic Risk Aversion Numerous experimental studies find that binary choice under risk is generally
probabilistic in nature (e.g. Camerer, 1989; Hey and Orme, 1994; Loomes and Sugden,
1998). In this section we will extend Definition 1 to a more general setup where people
may choose in a probabilistic manner. There are several alternative explanations why
people make inconsistent choices under risk when decision problems are repeated
within a short period of time (e.g. Loomes and Sugden, 1995). For example, people may
have multiple preference relations on ℒ or they may make random errors.
We will now assume that the primitive of choice is a binary choice probability
function P:ℒ ℒ→[0,1], which is also known as a fuzzy preference relation (e.g.
Zimmerman et al., 1984). The notation P(S,R) represents the probability that an
individual chooses lottery S ∈ ℒ over lottery R ∈ ℒ in a direct binary choice. For any
two lotteries S, R ∈ ℒ, S ≠ R, the probability P(S,R) is observable from the relative
7
frequency with which an individual chooses S when asked to choose repeatedly between
S and R. We will consider two individuals: an individual ♀ characterized by a binary
choice probability function P♀(.,.) and an individual ♂ characterized by a binary choice
probability function P♂(.,.).
Definition 1a An individual ♀ is probabilistically more risk averse than an
individual ♂ if P♀((x,1),L)≥P♂((x,1),L) for all x∈X and all L∈ℒ and there exists at least
one degenerate lottery (x,1)∈ℒ and one lottery L∈ℒ such that P♀((x,1),L)>P♂((x,1),L).
According to Definition 1a, a more risk averse individual is always at least as
likely to choose a degenerate lottery over a risky lottery as a less risk averse individual.
This definition of the more-risk-averse-than relation between individuals is very general.
As before, we do not restrict lottery outcomes to be measurable in real numbers. We
also do not require that fuzzy preferences over lotteries are represented by a specific
model of probabilistic choice. Thus, we can apply Definition 1a to very distinct models
of probabilistic choice, e.g. when people have multiple preference relations on ℒ
(Loomes and Sugden, 1995) or when people have a unique preference relation on ℒ but
they make random errors (Fechner, 1860; Hey and Orme, 1994; Blavatskyy, 2007).
If lottery L in Definition 1a is another degenerate lottery (y,1), y∈X, then we
arrive at the following result. If an individual ♀ is more risk averse than an individual
♂, or vice versa, then P♀((x,1), (y,1)) = P♂((x,1), (y,1)) for all x,y∈X. In other words, we
can unambiguously rank two individuals in terms of their risk attitudes only if they
choose in identical manner between riskless alternatives (degenerate lotteries). If this is
not the case, heterogeneous risk attitudes are confounded with heterogonous tastes over
riskless outcomes and we cannot make a clear comparison of individuals in terms or
relative risk aversion.
8
4. Risk Aversion in Models of Probabilistic Choice In this section we consider the implications of Definition 1a for several well-
known models of probabilistic choice. Arguably, the simplest model of probabilistic
choice is the constant error/tremble model of Harless and Camerer (1994). In this model,
an individual has a unique preference relation on ℒ but she does not always choose the
preferred lottery. With a constant probability τ∈[0,1] a tremble occurs and the individual
chooses the less preferred lottery. Specifically, in a constant error/tremble model there
exists an utility function u:X→ that is unique up to a linear transformation, such that
(3) P(S,R) = 0.5 + (0.5 – τ) sign( ∑x∈X S(x)u(x) – ∑x∈X R(x)u(x) ),
for any two lotteries S, R∈ℒ and a probability τ∈[0,1]. Formula (3) states that a lottery S
is chosen over a lottery R with probability 1-τ if the expected utility of S is greater than
the expected utility of R; with probability 0.5—if the expected utilities of lotteries S and
R are exactly equal; and with probability τ—if the expected utility of S is less than the
expected utility of R. The result of Proposition 1 can be extended to a constant
error/tremble model of probabilistic choice.
Proposition 2 If individual choices are represented by a constant error/tremble
model (3) then an individual ♀ with utility function u♀:X→ and the probability of a
tremble τ♀ is more risk averse than an individual ♂ with ordinally equivalent utility
function u♂:X→ and the probability of a tremble τ♂ = τ♀ if and only if
(4) ( ) ( )( ) ( )
( ) ( )( ) ( )
u y u x u y u xu z u y u z u y
− −≥
− −♀ ♀ ♂ ♂
♀ ♀ ♂ ♂
,
for any x, y, z ∈ X such that u♀(x) < u♀(y) < u♀(z) and there exists at least one triple of
outcomes {x, y, z} ⊂ X for which inequality (2) holds with strict inequality.
Proof is analogous to the proof of Proposition 1.
9
Not all models of probabilistic choice allow for an unambiguous ranking of
individuals in terms of their risk preferences. For example, consider a strong utility
model (e.g. Luce and Suppes, 1965). In a strong utility model there exists an utility
function u:X→ that is unique up to a positive linear transformation, and a strictly
increasing function :→[0,1], which satisfies (v)+(-v)=1 for all v∈, such that
(5) P(S,R) = ( ∑x∈X S(x)u(x) – ∑x∈X R(x)u(x) )
for any two lotteries S, R∈ℒ. If the function (.) is the cumulative distribution function
of a normal distribution with zero mean and constant standard deviation, model (5)
becomes the Fechner model of random errors (Fechner, 1860; Hey and Orme, 1994). If
the function (.) is the cumulative distribution function of the logistic distribution:
(v)=1/(1+exp(-λv)), where λ>0 is constant, model (5) becomes the Luce choice model
(Luce, 1959). Blavatskyy (2008) provides axiomatic characterization of model (5).
Proposition 3 If individual choices are represented by a strong utility model
(5) then it is impossible to find two individuals such that one of them is probabilistically
more risk averse than the other.
Proof is presented in the Appendix.
In a recent study, Wilcox (2008) discusses the failure of a strong utility model
to rank individuals in terms of their risk preferences (in the context of lotteries over
monetary outcomes).
10
5. Absolute Risk Aversion and Relative Riskiness So far we considered only comparative risk aversion. To measure absolute risk
aversion, we need to fix one binary choice probability function PRN : ℒ ℒ → [0,1]. An
individual is called risk neutral if she has the binary choice probability function PRN(.,.).
An individual is called risk averse if she is more risk averse (according to Definition 1a)
than the risk neutral individual. Similarly, an individual is called risk seeking or risk
loving if the risk neutral individual is more risk averse than this individual. Notice that
the concept of absolute risk aversion depends on an ad hoc selection of a risk neutral
binary choice probability function PRN(.,.). This is similar to our temperature
measurement that also requires an ad hoc selection of zero temperature (e.g. the triple
point of water in the Celsius scale or absolute zero in the Kelvin scale).
The concept of comparative risk aversion can be also used to define the
relative riskiness of lotteries. In a sense, we are now looking at the other side of a coin.
We ask the question: when is one lottery riskier than the other so that more risk averse
people dislike it? For expositional clarity, let us first define relative riskiness when
people have a unique rational preference relation on the set of lotteries ℒ.
One way to define relative riskiness is the following. A lottery R ∈ ℒ is riskier
than a lottery S ∈ ℒ if S ♂ R implies S ♀ R for any two individuals ♀ and ♂ such that
♀ is more risk averse than ♂, and S ♂ R implies S ♀ R for at least one such pair of
individuals. Aumann and Serrano (2007) use a similar definition of relative riskiness in
the context of lotteries over monetary outcomes.
However, there may exist two lotteries S, R ∈ ℒ such that every individual
strictly prefers S over R (in this case we say that lottery S dominates lottery R). For such
pair of lotteries, a more risk averse individual would always strictly prefer S over R.
11
However, this strong preference is not related to the relative riskiness of the two
lotteries. It simply reflects the fact that S is relatively better than R. To distinguish
between relative riskiness and relative attractiveness of lotteries, we use the following
definition.
Definition 2 A lottery R ∈ ℒ is riskier than a lottery S ∈ ℒ if S ≿♂ R implies
S ♀ R for any two individuals ♀ and ♂ such that ♀ is more risk averse than ♂, and
there exists at least one such pair of individuals for whom we have S ~♂ R and S ♀ R.
Definition 2 is more general than the traditional definitions of relative riskiness
in terms of second-order stochastic dominance or mean-preserving spreads (Rothschild
and Stiglitz, 1970). First of all, Definition 2 does not require lottery outcomes to be
measurable in real numbers. Second, traditional definitions of relative riskiness are
equivalent to Definition 2 with an additional restriction that an individual ♂ is risk-
neutral. Thus, the traditional definitions of relative riskiness apply only to a subset of
lotteries ℒ´ ⊂ ℒ such that for any S, R ∈ ℒ´ we have S ~RN R (when lottery outcomes are
monetary, this condition simply means that S and R have the same expected value). In
contrast, Definition 2 imposes a partial ordering in terms of relative riskiness on a
significantly larger set of lotteries.
Definition 2 defines riskiness as the attribute of lotteries that risk averse people
dislike. If we make additional assumptions about individual preferences over lotteries,
we can define riskiness in terms of objective characteristics of lotteries without referring
to subjective preferences. For example, let us consider individual preferences that are
represented by expected utility theory. As we already discussed above, comparative risk
aversion is well-defined under expected utility theory only if people have ordinally
equivalent utility functions.
12
Proposition 4 For a group of expected utility maximizers that have utility
functions ordinally equivalent to a function u:X→, a lottery R∈ℒ is riskier than another
lottery S ∈ ℒ if there exists an outcome y ∈ X such that ∑x∈X | u(x)≤v S(x) ≤ ∑x∈X | u(x)≤v R(x)
for any v < u(y) and ∑x∈X | u(x)≤w S(x) ≥ ∑x∈X | u(x)≤w R(x) for any w > u(y) and there exists at
least one v < u(y) and one w > u(y) such that both inequalities hold as strict inequalities.
Proof is presented in the Appendix.
Proposition 4 can be conveniently illustrated inside the probability triangle
when the outcome set is X={x, y, z}. As it is conventional, we plot the probability of the
outcome x (z) that yields the lowest (highest) utility on the horizontal (vertical) axis.
Figure 2 shows the set {R∈ℒ|R(x)>L(x),R(z)>L(z)} of all lotteries that are riskier than an
arbitrary selected lottery L and the set of all lotteries {S∈ℒ|S(x)<L(x),S(z)<L(z)} that are
less risky than L. Figure 2 adheres to informal usage of the concept of relative riskiness.
Many studies refer to the lotteries located in the north-eastern direction as “riskier” and
to the lotteries located in the south-western direction—as “safer” but do not provide a
formal definition of relative riskiness (e.g. Camerer, 1989; Loomes and Sugden, 1998).
Figure 2 The set of all lotteries that are riskier than lottery L and the set of all lotteries that are less risky than lottery L: illustration in the probability triangle
L(z)
L(x) 1
Probability of outcome z
Probability of outcome x
Set of all lotteries that are less risky than L {S∈ℒ|S(x)<L(x),S(z)<L(z)}
1
L
0
Set of all lotteries that are riskier than L {R∈ℒ|R(x)>L(x),R(z)>L(z)}
13
We can extend Definition 2 to a more general setup where people have fuzzy
preferences captured by a binary choice probability function P:ℒ ℒ → [0,1].
Definition 2a A lottery R∈ℒ is riskier than a lottery S∈ℒ if P♀(S,R)≥P♂(S,R)
for any two individuals ♀ and ♂ such that ♀ is more risk averse than ♂, and there exists
at least one such pair of individuals for whom we have P♀(S,R)>P♂(S,R).
6. Conclusion Risk aversion is a fundamental concept in many fields of economics. However,
it is traditionally defined only in the context of lotteries over monetary payoffs. This
paper extends the definition of risk aversion to a more general setup where lottery
outcomes are not necessarily measurable in real numbers and people do not necessarily
have a unique preference relation over risky lotteries, i.e. they may choose in a
probabilistic manner.
We show that in neoclassical expected utility theory risk aversion can be
captured by a simple index of local risk aversion (the slope of indifference curves inside
the “local” probability triangle). The same result holds for a constant error/ tremble
model of probabilistic choice. However, not all models of probabilistic choice allow for
an unambiguous ranking of individuals in terms of their risk preferences. In particular,
we prove an impossibility theorem for a strong utility model of probabilistic choice
(which includes the Fechner model and the Luce choice model as special cases).
Finally, we show that the definition of comparative risk aversion can be used
to define a related concept of the relative riskiness of lotteries. Our proposed definition
of relative riskiness generalizes traditional definitions (second-order stochastic
dominance, mean preserving spreads) to a larger class of lotteries (that may differ in
expected value or may yield non-monetary payoffs). The proposed definition adheres to
the informal usage of the concept of relative riskiness in the literature.
14
References Aumann, R. and R. Serrano (2007) "An Economic Index of Riskiness," Discussion
Paper Series dp446, Center for Rationality and Interactive Decision Theory,
Hebrew University, Jerusalem
Blavatskyy, P. (2008) “Stochastic Utility Theorem” Journal of Mathematical
Economics, forthcoming
Blavatskyy, Pavlo (2007) “Stochastic Expected Utility Theory” Journal of Risk and
Uncertainty 34, 259-286
Camerer, C. (1989) “An experimental test of several generalized utility theories.”
Journal of Risk and Uncertainty 2, 61-104
Epstein, L. (1999) “A Definition of Uncertainty Aversion” Review of Economic Studies
66, 579-608
Fechner, G. (1860) “Elements of Psychophysics” NewYork: Holt, Rinehart and
Winston
Harless, D. and C. Camerer (1994) The predictive utility of generalized expected utility
theories, Econometrica 62, 1251-1289
Hey, J.D. and C. Orme (1994) Investigating generalisations of expected utility theory
using experimental data, Econometrica 62, 1291-1326
Hilton, R. (1989) “Risk Attitude under Random Utility” Journal of Mathematical
Psychology 33, 206-222
Loomes, G. and Sugden, R. (1995) “Incorporating a stochastic element into decision
theories” European Economic Review 39, 641-648
Loomes, G. and Sugden, R. (1998) “Testing different stochastic specifications of risky
choice” Economica 65, 581-598
15
Luce, R. D. (1959) “Individual choice behavior” New York: Wiley
Luce, R. D. and Suppes, P. (1965) “Preference, utility, and subjective probability” in R.
D. Luce, R. R. Bush & E. Galanter (eds.), Handbook of mathematical
psychology, Vol. III, 249–410, Wiley, New York NY
Machina, M. (1982) “Expected utility’ analysis without the independence axiom”,
Econometrica 50, 277–323
Pratt, J. (1964) “Risk Aversion in the Small and in the Large” Econometrica 66, 122-
136
Rothschild, M. and J. Stiglitz (1970) “Increasing risk I: a definition” Journal of
Economic Theory 2, 225-243
von Neumann, J. and Morgenstern, O. (1944) “Theory of Games and Economic
Behavior, ” Princeton, Princeton University Press
Wilcox, N. (2008) “’Stochastically More Risk Averse’: A Contextual Theory of
Stochastic Discrete Choice under Risk” Journal of Econometrics forthcoming
Zimmerman, H.J., Gaines, B.R. and Zadeh, L. A. (1984) “Fuzzy Sets and Decision
Analysis”, North-Holland
16
Appendix Proof of Proposition 1.
Let us first prove the necessity of condition (2). If an individual ♀ is more risk
averse than an individual ♂ then there exists at least one triple of outcomes {x, y, z}⊂X
such that u♀(x) < u♀(y) < u♀(z). Otherwise, an utility function u♀:X→ maps all
outcomes to only one or two real numbers and an ordinally equivalent utility function
u♂:X→ does the same. Hence, utility function u♂:X→ is a linear transformation of
utility function u♀:X→ and both individuals have the same binary preference relation
on ℒ i.e. an individual ♀ cannot be more risk averse than an individual ♂.
For any triple {x, y, z}⊂X such that u♀(x) < u♀(y) < u♀(z) we can construct a
lottery L that yields an outcome x with a probability 1-q and an outcome z with a
probability q∈(0,1). An expected utility maximizer ♂ prefers a degenerate lottery that
yields outcome y for sure over lottery L if u♂(y) ≥ (1-q)u♂(x) + qu♂(z). We can rearrange
this condition into
(6) ( ) ( )( ) ( ) 1
u y u x qu z u y q
−≥
− −♂ ♂
♂ ♂
.
Similarly, an expected utility maximizer ♀ prefers a degenerate lottery that
yields outcome y for sure over lottery L if
(7) ( ) ( )( ) ( ) 1
u y u x qu z u y q
−≥
− −♀ ♀
♀ ♀
.
If the left-hand-side of (7) is strictly less than the left-hand-side of (6), then we can find
a probability q∈(0,1) sufficiently close to one such that inequality (6) holds but
inequality (7) does not hold. However, this contradicts to our premise that an individual
♀ is more risk averse than an individual ♂ (we found a degenerate lottery and a risky
17
lottery L such that an individual ♂ prefers the degenerate lottery over L but an
individual ♀ does not). Hence, the left-hand-side of (7) should be greater than or equal
to the left-hand-side of (6) for any triple {x, y, z}⊂X such that u♀(x) < u♀(y) < u♀(z).
Let us now prove that condition (2) is sufficient for characterizing an
individual ♀ as more risk averse. An individual ♂ prefers a degenerate lottery that
yields an outcome y ∈ X for sure over an arbitrary lottery L ∈ ℒ if
(8) u♂(y) ≥ ∑x∈X L(x)u♂(x).
If u♂(y) = maxx∈X u♂(x) then condition (8) is satisfied for any lottery L. Since individuals
♀ and ♂ have ordinally equivalent utility functions, it must be also the case that u♀(y) =
maxx∈X u♀(x) so that u♀(y) ≥ ∑x∈X L(x)u♀(x). Thus, an individual ♀ also prefers a
degenerate lottery that yields y for sure over any lottery L.
If u♂(y) = minx∈X u♂(x) then condition (8) can be satisfied only if lottery L
yields the lowest possible expected utility u♂(y). Since individuals ♀ and ♂ have
ordinally equivalent utility functions, it must be also the case that u♀(y) = minx∈X u♀(x)=
=∑x∈X L(x)u♀(x). Thus, in this case, if an individual ♂ weakly prefers a degenerate
lottery that yields y for sure over lottery L, an individual ♀ does so as well.
If u♂(y) ≠ maxx∈X u♂(x) and u♂(y) ≠ minx∈X u♂(x) then it is possible to find an
outcome w ∈ X that has the highest utility u♂(w) such that u♂(w) < u♂(y). Similarly, it is
possible to find an outcome z ∈ X that has the lowest utility u♂(z) so that u♂(z)>u♂(y). For
convenience, let us introduce the following notation. For an arbitrary lottery L ∈ ℒ let
q(w) denote the cumulative probability of all outcomes that have the utility of u♂(w), i.e.
q(w) ≡ ∑x∈X| u♂(x)=u♂(w) L(x). Similarly, let us define q(y) ≡ ∑x∈X| u♂(x)=u♂(y) L(x) and q(z) ≡
18
∑x∈X| u♂(x)=u♂(z) L(x). An individual ♂ prefers a degenerate lottery that yields an outcome
y for sure over an arbitrary lottery L ∈ ℒ if
(9) ( ) ( ) ( )( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( )( ) ( ) .x X u x u w
x X u x u z
u y L x u x q w u w q y u y
q z u z L x u x∈ <
∈ >
≥ + + +
+ +
∑∑
♂ ♂
♂ ♂
♂ ♂ ♂ ♂
♂ ♂
We can rearrange inequality (9) into the following condition
(10)
( ) ( )( ) ( ) ( )( ) ( ) ( ) ( )
( ) ( )( ) ( )
( ) ( ) ( )( ) ( )( ) ( ) ( )
1
.
x X u x u w
x X u x u z
u y u w u w u xq y L x
u z u w u z u w
u x u wL x q z
u z u w
∈ <
∈ >
− −− + −
− −
−− ≥
−
∑
∑
♂ ♂
♂ ♂
♂ ♂ ♂ ♂
♂ ♂ ♂ ♂
♂ ♂
♂ ♂
If condition (2) holds then we have ( ) ( )( ) ( )
( ) ( )( ) ( )
u w u x u w u xu z u w u z u w
− −≥
− −♀ ♀ ♂ ♂
♀ ♀ ♂ ♂
for any
outcome x∈X such that u♂(x)<u♂(w). Condition (2) also implies that
( ) ( )( ) ( )
( ) ( )( ) ( )
u y u w u y u wu z u w u z u w
− −≥
− −♀ ♀ ♂ ♂
♀ ♀ ♂ ♂
and ( ) ( )( ) ( )
( ) ( )( ) ( )
u x u w u x u wu z u w u z u w
− −− ≥ −
− −♀ ♀ ♀ ♀
♀ ♀ ♂ ♂
for any
outcome x∈X such that u♂(x)>u♂(z). Using these results and inequality (10), we can write
(11)
( ) ( )( ) ( ) ( )( ) ( ) ( ) ( )
( ) ( )( ) ( )
( ) ( ) ( )( ) ( )( ) ( ) ( )
1
.
x X u x u w
x X u x u z
u y u w u w u xq y L x
u z u w u z u w
u x u wL x q z
u z u w
∈ <
∈ >
− −− + −
− −
−− ≥
−
∑
∑
♂ ♂
♂ ♂
♀ ♀ ♀ ♀
♀ ♀ ♀ ♀
♀ ♀
♀ ♀
Finally, inequality (11) can be rewritten as u♀(y) ≥ ∑x∈X L(x)u♀(x). To sum up,
if an individual ♂ prefers a degenerate lottery that yields an arbitrary outcome y for sure
over an arbitrary lottery L and condition (2) holds, then an individual ♀ also prefers the
degenerate lottery that yields outcome y for sure over lottery L. Thus, according to
Definition 1, an individual ♀ is more risk averse than an individual ♂. Q.E.D.
19
Proof of Proposition 3.
Suppose that there is an individual ♀, characterized by utility function
u♀:X→ and function ♀:→[0,1], who is more risk averse than another individual ♂,
characterized by utility function u♂:X→ and function ♂:→[0,1]. First, let us prove
that there is a constant k such that ♀(v) = ♂(kv) for all v ∈ .
Let y ∈ X be an outcome such that u♂(y) = minx∈X u♂(x) and let z ∈ X be an
outcome such that u♂(z) = maxx∈X u♂(x). Note that if u♂(y) = u♂(z) then an individual ♂
chooses with probabilities 50%-50% between any two lotteries (including degenerate
lotteries). Since both individuals should have identical binary choice probabilities in
choice under certainty, it follows that an individual ♀ also chooses with probabilities
50%-50% between any two degenerate lotteries. This implies that function u♀:X→
maps all outcomes to just one number and individual ♀ also chooses with probabilities
50%-50% between any two risky lotteries i.e. she cannot be more risk averse than an
individual ♂. Therefore, we need only to consider the case when u♂(y) > u♂(z).
Let us consider a degenerate lottery (y,1) that yields outcome y for sure and a
risky lottery L that yields outcome y with probability 1-q and outcome z with probability
q∈[0,1]. An individual ♂ chooses (y,1) over L with a probability ♂(-q(u♂(z)-u♂(y))). An
individual ♀ chooses (y,1) over L with a probability ♀(-q(u♀(z)- u♀(y))). If an individual
♀ is more risk averse than and individual ♂ we must have
(12) ♀(-q(u♀(z)- u♀(y))) ≥ ♂(-q(u♂(z)-u♂(y))).
Since ♀(-v) = 1 - ♀(v) and ♂(-v) = 1 - ♂(v) for any v ∈ we can rewrite (12) as
(13) ♂(q(u♂(z)-u♂(y))) ≥ ♀(q(u♀(z)- u♀(y))).
Let us consider a degenerate lottery (z,1) that yields outcome z for sure and a
risky lottery L´ that yields outcome y with probability q and outcome z with probability
20
1-q. An individual ♂ chooses (z,1) over L´ with a probability ♂(q(u♂(z)-u♂(y))). An
individual ♀ chooses (z,1) over L´ with a probability ♀(q(u♀(z)- u♀(y))). If an individual
♀ is more risk averse than and individual ♂ we must have
(14) ♀(q(u♀(z)- u♀(y))) ≥ ♂(q(u♂(z)-u♂(y))).
Inequalities (13) and (14) can hold simultaneously only if
(15) ♀(q(u♀(z)- u♀(y))) = ♂(q(u♂(z)-u♂(y)))
for any q∈[0,1]. Using a substitution of variables q = v/(u♀(z)- u♀(y)) in equation (15) we
arrive at ♀(v) = ♂(kv), were k = (u♂(z)-u♂(y))/(u♀(z)- u♀(y)) >0 is constant.
Let us now prove that utility function u♀(.) is a positive linear transformation
of utility function u♂(.) i.e. there are constants a>0 and b such that u♀(x) = au♂(x) + b for
all x∈X. An individual ♂ chooses a degenerate lottery (x,1) over a degenerate lottery
(y,1) with a probability ♂(u♂(x)-u♂(y)). An individual ♀ chooses (x,1) over (y,1) with a
probability ♀(u♀(x)- u♀(y)). If the two individuals can be ranked in terms of their risk
preferences, we must have
(16) ♀(u♀(x)- u♀(y)) = ♂(u♂(x)-u♂(y)).
We already established that ♀(u♀(x)- u♀(y)) = ♂(k(u♀(x)- u♀(y))). Plugging this
result into (16) we receive
(17) ♂(k(u♀(x)- u♀(y))) = ♂(u♂(x)-u♂(y)).
Since function ♂:→[0,1] is strictly increasing, equation (17) holds only if
(18) k(u♀(x)- u♀(y)) = u♂(x)-u♂(y).
Rearranging (18) we obtain u♀(x) = u♂(x)/k + u♀(y) - u♂(y)/k i.e. utility function
u♀(.) is a positive linear transformation of utility function u♂(.). Thus, individuals ♀ and
♂ choose in an identical manner between any two lotteries. In other words, an
individual ♀ cannot be more risk averse than an individual ♂. Q.E.D.
21
Proof of Proposition 4.
Let U⊂ be the range (the image of the domain) of an utility function u:X→.
Since the outcome set X is a finite nonempty set, U must be also a finite nonempty set of
real numbers. Thus, we can number the elements of U so that ui<uj whenever i<j for any
ui, uj ∈U and any i, j ∈ {1,…,|U|}. For any lottery S∈ℒ and any number ui∈U let Śi denote
the cumulative probability of all outcomes that yield utility ui, i.e. Śi ≡ ∑x∈X| u(x)=ui S(x).
Let us now consider an expected utility maximizer ♀ with utility function
u♀:X→ who is more risk averse than an expected utility maximizer ♂ with ordinally
equivalent utility function u♂:X→. An individual ♂ weakly prefers lottery S over
lottery R if and only if ∑x∈X [S(x)-R(x)]u♂(x) ≥ 0, which is equivalent to:
(19) 1
0U ii iiŚ Ŕ u
=⎡ ⎤− ≥⎣ ⎦∑ ♂ .
Inequality (19) can be then rearranged into:
(20) ( )( ) ( )( )1 1 11 1 1
0k j U Uk j j j ji i i ij i j k i j
u Ŕ Ś u u Ś Ŕ u u− + −= = = + =
⎡ ⎤ ⎡ ⎤+ − − + − − ≥⎣ ⎦ ⎣ ⎦∑ ∑ ∑ ∑♂ ♂ ♂ ♂ ♂ ,
for any k∈{2,…,|U|-1}. Furthermore, we can rewrite inequality (20) as follows:
(21)
2 1 3 21
1 1 1 2 1 23 2 4 3 1
11
1 11 1 2
1 2
2 3
ki ii
U Uk k
U U U U U Uk k U U
U U
iU U
u u u uŔ Ś Ŕ Ŕ Ś Ś Ŕ Śu u u u
u u u uŚ Ŕ Ś Ś Ŕ Ŕu u u u
u u Śu u
−
=
−−
− −+ − −
− −
− −
⎛ ⎞⎛ ⎞− −⎡ ⎤ ⎡ ⎤ ⎡ ⎤− + + − + + + − ×⎜ ⎟⎜ ⎟⎣ ⎦ ⎣ ⎦ ⎣ ⎦⎜ ⎟− −⎝ ⎠⎝ ⎠⎛ ⎛ ⎞− −⎡ ⎤ ⎡ ⎤× + − + + − − ×⎜ ⎜ ⎟⎣ ⎦ ⎣ ⎦⎜ ⎟⎜− −⎝ ⎠⎝
−× + + −
−
∑… …
…
…
♂ ♂ ♂ ♂
♂ ♂ ♂ ♂
♂ ♂ ♂ ♂
♂ ♂ ♂ ♂
♂ ♂
♂ ♂11
0.k
Ui k ki k
uŔu u+= +
⎞⎡ ⎤ + ≥⎟⎣ ⎦ ⎟ −⎠
∑ ♂
♂ ♂
According to Proposition 1, if an expected utility maximizer ♀ is more risk
averse than an expected utility maximizer ♂ then we must have:
(22) j i j i
m j m j
u u u uu u u u
− −≥
− −♀ ♀ ♂ ♂
♀ ♀ ♂ ♂
22
for any i, j, m ∈ {1,…,|U|} such that i < j < m. Note that if there exists a number
k∈{2,…,|U|-1} such that 1
0ji iiŔ Ś
=⎡ ⎤− ≥⎣ ⎦∑ for any j∈{1,…,k-1} and 0U
i ii jŚ Ŕ
=⎡ ⎤− ≤⎣ ⎦∑
for any j∈{k+1,…, |U|} then we can use inequality (22) to rewrite (21) as:
(23)
2 1 3 21
1 1 1 2 1 23 2 4 3 1
11
1 11 1 2
1 2
2 3
ki ii
U Uk k
U U U U U Uk k U U
U U
iU U
u u u uŔ Ś Ŕ Ŕ Ś Ś Ŕ Ś
u u u u
u u u uŚ Ŕ Ś Ś Ŕ Ŕ
u u u u
u uŚ
u u
−
=
−−
− −+ − −
− −
− −
⎛ ⎞⎛ ⎞− −⎡ ⎤ ⎡ ⎤ ⎡ ⎤− + + − + + + − ×⎜ ⎟⎜ ⎟⎣ ⎦ ⎣ ⎦ ⎣ ⎦⎜ ⎟− −⎝ ⎠⎝ ⎠
⎛ ⎛ ⎞− −⎡ ⎤ ⎡ ⎤× + − + + − − ×⎜ ⎜ ⎟⎣ ⎦ ⎣ ⎦⎜ ⎟⎜− −⎝ ⎠⎝
−× + + −
−
∑… …
…
…
♀ ♀ ♀ ♀
♀ ♀ ♀ ♀
♀ ♀ ♀ ♀
♀ ♀ ♀ ♀
♀ ♀
♀ ♀11
0.k
Ui k ki k
uŔ
u u+= +
⎞⎡ ⎤ + ≥⎟⎣ ⎦ ⎟ −⎠
∑ ♀
♀ ♀
Finally, we can rearrange (23) into 1
0U ii iiŚ Ŕ u
=⎡ ⎤− ≥⎣ ⎦∑ ♀ , which is equivalent to:
(24) ∑x∈X [S(x)-R(x)] u♀(x) ≥ 0.
To summarize, we showed that a more risk averse expected utility maximizer
♀ weakly prefers lottery S over lottery R whenever a less risk averse expected utility
maximizer ♂ weakly prefers S over R, if there exists a number k∈{2,…,|U|-1} such that
10j
i iiŔ Ś
=⎡ ⎤− ≥⎣ ⎦∑ for any j∈{1,…,k-1} and 0U
i ii jŚ Ŕ
=⎡ ⎤− ≤⎣ ⎦∑ for any j∈{k+1,…, |U|}.
Note that if 0Ui ii jŚ Ŕ
=⎡ ⎤− =⎣ ⎦∑ for all j∈{k+1,…, |U|} then the left-hand side
of (21) is strictly positive, i.e. we cannot find an individual ♂ who is exactly indifferent
between S and R. In this case, according to Definition 2, S cannot be riskier than R.
Therefore, we must have 0Ui ii jŚ Ŕ
=⎡ ⎤− <⎣ ⎦∑ for at least one j∈{k+1,…, |U|}. By a
similar argument, we must also have 1
0ji iiŔ Ś
=⎡ ⎤− >⎣ ⎦∑ for at least one j∈{1,…,k-1}.
Hence, lottery R is riskier than lottery S if there exists an outcome y ∈ X such
that ∑x∈X | u(x)≤v S(x) ≤ ∑x∈X | u(x)≤v R(x) for any v < u(y) and ∑x∈X | u(x)≤w S(x) ≥ ∑x∈X | u(x)≤w R(x)
for any w > u(y) and there exists at least one v < u(y) and one w > u(y) such that both
inequalities hold as strict inequalities. Q.E.D.
